Collaborative Research: Algorithms for Optimal Adaptive Enrichment Design in Randomized Trial

临床试验费用的增加降低了制药公司进行临床试验的意愿，推迟了新药开发，从而对公共卫生产生了负面影响。 一个典型的临床试验可能会花费数百万美元，这取决于治疗领域和科学目标。在设计临床试验时，需要在科学/生物学方面，统计功效和成本方面平衡（通常是相互冲突的）目标。在个性化医疗中，这个决策问题非常复杂，需要同时考虑多个亚群。现有的设计适应性试验的方法要么不涉及目标优化，要么在非常严格的设置中进行优化。该项目考虑自适应富集设计，一种灵活的试验设计框架，允许试验管理员在试验期间调整患者入组规则。它已被证明经常提供上级成本效益和更好的统计能力。本研究旨在设计新的方法和算法来优化自适应富集设计，将具有两个规划阶段和两个子种群的优化设计问题表示为一个大规模线性规划模型，可以用现成的LP求解器求解。由于线性规划的规模呈指数增长，这种线性规划求解器不能直接应用于具有更多规划阶段和子种群的实际情况。该项目将开发专门的算法和建模技术，以充分利用问题结构，更快地解决两阶段两子群模型，并将其扩展到以前被认为无法解决的更大模型。此外，还将开发方便用户的开放源码软件，使科学家能够自行设计最佳的适应性浓缩设计，这一奖项反映了国家科学基金会的法定使命，并被认为值得通过使用基金会的知识价值和更广泛的影响审查标准进行评估而获得支持。

The increasing costs of clinical trials negatively impact public health by reducing drug companies' willingness to undertake clinical trials and delaying new drug development. A typical clinical trial may cost millions of U.S. dollars, depending on therapeutic areas and scientific goals. In designing clinical trials, one needs to balance (typically conflicting) aims in scientific/biological aspects, statistical power, and cost. This decision-making problem is very complicated in personalized medicine, where multiple subpopulations need to be simultaneously considered. Existing approaches for designing adaptive trials either do not involve optimization of objectives or optimize in very restrictive settings. This project considers adaptive enrichment design, a flexible trial design framework that allows trial administrators to adjust patient enrollment rules during the trials. It has been shown to often provide superior cost effectiveness and better statistical power. The research aims to design new methods and algorithms to optimize adaptive enrichment design.The optimal design problem with two planning stages and two subpopulations is formulated as a large-scale linear programming model, which can be solved by off-the-shelf LP solvers. Due to the exponentially increasing LP size, such LP solvers cannot be directly applied to the practical situations with more planning stages and subpopulations. This project will develop specialized algorithms and modelling techniques to fully exploit problem structures to solve two-stage two-subpopulation models much faster, and extend them to larger models previously regarded as unsolvable. Furthermore, user-friendly open-source software will be developed to enable scientists to construct their own optimal adaptive enrichment designs.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.